summaryrefslogtreecommitdiff
path: root/compiler/optimizing
diff options
context:
space:
mode:
Diffstat (limited to 'compiler/optimizing')
-rw-r--r--compiler/optimizing/code_generator.cc24
-rw-r--r--compiler/optimizing/code_generator.h22
-rw-r--r--compiler/optimizing/optimizing_cfi_test.cc151
-rw-r--r--compiler/optimizing/optimizing_cfi_test_expected.inc275
-rw-r--r--compiler/optimizing/optimizing_compiler.cc346
5 files changed, 613 insertions, 205 deletions
diff --git a/compiler/optimizing/code_generator.cc b/compiler/optimizing/code_generator.cc
index a1bb5e0838..ce92470868 100644
--- a/compiler/optimizing/code_generator.cc
+++ b/compiler/optimizing/code_generator.cc
@@ -42,7 +42,7 @@
#include "compiled_method.h"
#include "dex/verified_method.h"
-#include "driver/dex_compilation_unit.h"
+#include "driver/compiler_driver.h"
#include "gc_map_builder.h"
#include "graph_visualizer.h"
#include "intrinsics.h"
@@ -787,9 +787,10 @@ CodeGenerator* CodeGenerator::Create(HGraph* graph,
}
void CodeGenerator::BuildNativeGCMap(
- ArenaVector<uint8_t>* data, const DexCompilationUnit& dex_compilation_unit) const {
+ ArenaVector<uint8_t>* data, const CompilerDriver& compiler_driver) const {
const std::vector<uint8_t>& gc_map_raw =
- dex_compilation_unit.GetVerifiedMethod()->GetDexGcMap();
+ compiler_driver.GetVerifiedMethod(&GetGraph()->GetDexFile(), GetGraph()->GetMethodIdx())
+ ->GetDexGcMap();
verifier::DexPcToReferenceMap dex_gc_map(&(gc_map_raw)[0]);
uint32_t max_native_offset = stack_map_stream_.ComputeMaxNativePcOffset();
@@ -911,19 +912,22 @@ void CodeGenerator::BuildVMapTable(ArenaVector<uint8_t>* data) const {
vmap_encoder.PushBackUnsigned(VmapTable::kAdjustedFpMarker);
}
-void CodeGenerator::BuildStackMaps(ArenaVector<uint8_t>* data) {
- uint32_t size = stack_map_stream_.PrepareForFillIn();
- data->resize(size);
- MemoryRegion region(data->data(), size);
+size_t CodeGenerator::ComputeStackMapsSize() {
+ return stack_map_stream_.PrepareForFillIn();
+}
+
+void CodeGenerator::BuildStackMaps(MemoryRegion region) {
stack_map_stream_.FillIn(region);
}
void CodeGenerator::RecordNativeDebugInfo(uint32_t dex_pc,
uintptr_t native_pc_begin,
uintptr_t native_pc_end) {
- if (src_map_ != nullptr && dex_pc != kNoDexPc && native_pc_begin != native_pc_end) {
- src_map_->push_back(SrcMapElem({static_cast<uint32_t>(native_pc_begin),
- static_cast<int32_t>(dex_pc)}));
+ if (compiler_options_.GetGenerateDebugInfo() &&
+ dex_pc != kNoDexPc &&
+ native_pc_begin != native_pc_end) {
+ src_map_.push_back(SrcMapElem({static_cast<uint32_t>(native_pc_begin),
+ static_cast<int32_t>(dex_pc)}));
}
}
diff --git a/compiler/optimizing/code_generator.h b/compiler/optimizing/code_generator.h
index 47b6f30450..a92014dc79 100644
--- a/compiler/optimizing/code_generator.h
+++ b/compiler/optimizing/code_generator.h
@@ -22,6 +22,7 @@
#include "base/arena_containers.h"
#include "base/arena_object.h"
#include "base/bit_field.h"
+#include "compiled_method.h"
#include "driver/compiler_options.h"
#include "globals.h"
#include "graph_visualizer.h"
@@ -51,13 +52,9 @@ static int64_t constexpr kPrimLongMax = INT64_C(0x7fffffffffffffff);
class Assembler;
class CodeGenerator;
-class DexCompilationUnit;
+class CompilerDriver;
class LinkerPatch;
class ParallelMoveResolver;
-class SrcMapElem;
-template <class Alloc>
-class SrcMap;
-using DefaultSrcMap = SrcMap<std::allocator<SrcMapElem>>;
class CodeAllocator {
public:
@@ -284,13 +281,12 @@ class CodeGenerator {
slow_paths_.push_back(slow_path);
}
- void SetSrcMap(DefaultSrcMap* src_map) { src_map_ = src_map; }
-
void BuildMappingTable(ArenaVector<uint8_t>* vector) const;
void BuildVMapTable(ArenaVector<uint8_t>* vector) const;
void BuildNativeGCMap(
- ArenaVector<uint8_t>* vector, const DexCompilationUnit& dex_compilation_unit) const;
- void BuildStackMaps(ArenaVector<uint8_t>* vector);
+ ArenaVector<uint8_t>* vector, const CompilerDriver& compiler_driver) const;
+ void BuildStackMaps(MemoryRegion region);
+ size_t ComputeStackMapsSize();
bool IsBaseline() const {
return is_baseline_;
@@ -446,6 +442,10 @@ class CodeGenerator {
// Copy the result of a call into the given target.
virtual void MoveFromReturnRegister(Location trg, Primitive::Type type) = 0;
+ const ArenaVector<SrcMapElem>& GetSrcMappingTable() const {
+ return src_map_;
+ }
+
protected:
// Method patch info used for recording locations of required linker patches and
// target methods. The target method can be used for various purposes, whether for
@@ -488,7 +488,7 @@ class CodeGenerator {
stats_(stats),
graph_(graph),
compiler_options_(compiler_options),
- src_map_(nullptr),
+ src_map_(graph->GetArena()->Adapter(kArenaAllocCodeGenerator)),
slow_paths_(graph->GetArena()->Adapter(kArenaAllocCodeGenerator)),
current_block_index_(0),
is_leaf_(true),
@@ -602,7 +602,7 @@ class CodeGenerator {
const CompilerOptions& compiler_options_;
// Native to dex_pc map used for native debugging/profiling tools.
- DefaultSrcMap* src_map_;
+ ArenaVector<SrcMapElem> src_map_;
ArenaVector<SlowPathCode*> slow_paths_;
// The current block index in `block_order_` of the block
diff --git a/compiler/optimizing/optimizing_cfi_test.cc b/compiler/optimizing/optimizing_cfi_test.cc
index 05c6b2ca57..34f1fe5949 100644
--- a/compiler/optimizing/optimizing_cfi_test.cc
+++ b/compiler/optimizing/optimizing_cfi_test.cc
@@ -23,6 +23,8 @@
#include "optimizing/code_generator.h"
#include "optimizing/optimizing_unit_test.h"
#include "utils/assembler.h"
+#include "utils/arm/assembler_thumb2.h"
+#include "utils/mips/assembler_mips.h"
#include "optimizing/optimizing_cfi_test_expected.inc"
@@ -36,52 +38,62 @@ class OptimizingCFITest : public CFITest {
// Enable this flag to generate the expected outputs.
static constexpr bool kGenerateExpected = false;
- void TestImpl(InstructionSet isa, const char* isa_str,
- const std::vector<uint8_t>& expected_asm,
- const std::vector<uint8_t>& expected_cfi) {
+ OptimizingCFITest()
+ : pool_(),
+ allocator_(&pool_),
+ opts_(),
+ isa_features_(),
+ graph_(nullptr),
+ code_gen_(),
+ blocks_(allocator_.Adapter()) {}
+
+ void SetUpFrame(InstructionSet isa) {
// Setup simple context.
- ArenaPool pool;
- ArenaAllocator allocator(&pool);
- CompilerOptions opts;
- std::unique_ptr<const InstructionSetFeatures> isa_features;
std::string error;
- isa_features.reset(InstructionSetFeatures::FromVariant(isa, "default", &error));
- HGraph* graph = CreateGraph(&allocator);
+ isa_features_.reset(InstructionSetFeatures::FromVariant(isa, "default", &error));
+ graph_ = CreateGraph(&allocator_);
// Generate simple frame with some spills.
- std::unique_ptr<CodeGenerator> code_gen(
- CodeGenerator::Create(graph, isa, *isa_features.get(), opts));
+ code_gen_.reset(CodeGenerator::Create(graph_, isa, *isa_features_, opts_));
+ code_gen_->GetAssembler()->cfi().SetEnabled(true);
const int frame_size = 64;
int core_reg = 0;
int fp_reg = 0;
for (int i = 0; i < 2; i++) { // Two registers of each kind.
for (; core_reg < 32; core_reg++) {
- if (code_gen->IsCoreCalleeSaveRegister(core_reg)) {
+ if (code_gen_->IsCoreCalleeSaveRegister(core_reg)) {
auto location = Location::RegisterLocation(core_reg);
- code_gen->AddAllocatedRegister(location);
+ code_gen_->AddAllocatedRegister(location);
core_reg++;
break;
}
}
for (; fp_reg < 32; fp_reg++) {
- if (code_gen->IsFloatingPointCalleeSaveRegister(fp_reg)) {
+ if (code_gen_->IsFloatingPointCalleeSaveRegister(fp_reg)) {
auto location = Location::FpuRegisterLocation(fp_reg);
- code_gen->AddAllocatedRegister(location);
+ code_gen_->AddAllocatedRegister(location);
fp_reg++;
break;
}
}
}
- ArenaVector<HBasicBlock*> blocks(allocator.Adapter());
- code_gen->block_order_ = &blocks;
- code_gen->ComputeSpillMask();
- code_gen->SetFrameSize(frame_size);
- code_gen->GenerateFrameEntry();
- code_gen->GenerateFrameExit();
+ code_gen_->block_order_ = &blocks_;
+ code_gen_->ComputeSpillMask();
+ code_gen_->SetFrameSize(frame_size);
+ code_gen_->GenerateFrameEntry();
+ }
+
+ void Finish() {
+ code_gen_->GenerateFrameExit();
+ code_gen_->Finalize(&code_allocator_);
+ }
+
+ void Check(InstructionSet isa,
+ const char* isa_str,
+ const std::vector<uint8_t>& expected_asm,
+ const std::vector<uint8_t>& expected_cfi) {
// Get the outputs.
- InternalCodeAllocator code_allocator;
- code_gen->Finalize(&code_allocator);
- const std::vector<uint8_t>& actual_asm = code_allocator.GetMemory();
- Assembler* opt_asm = code_gen->GetAssembler();
+ const std::vector<uint8_t>& actual_asm = code_allocator_.GetMemory();
+ Assembler* opt_asm = code_gen_->GetAssembler();
const std::vector<uint8_t>& actual_cfi = *(opt_asm->cfi().data());
if (kGenerateExpected) {
@@ -92,6 +104,19 @@ class OptimizingCFITest : public CFITest {
}
}
+ void TestImpl(InstructionSet isa, const char*
+ isa_str,
+ const std::vector<uint8_t>& expected_asm,
+ const std::vector<uint8_t>& expected_cfi) {
+ SetUpFrame(isa);
+ Finish();
+ Check(isa, isa_str, expected_asm, expected_cfi);
+ }
+
+ CodeGenerator* GetCodeGenerator() {
+ return code_gen_.get();
+ }
+
private:
class InternalCodeAllocator : public CodeAllocator {
public:
@@ -109,21 +134,83 @@ class OptimizingCFITest : public CFITest {
DISALLOW_COPY_AND_ASSIGN(InternalCodeAllocator);
};
+
+ ArenaPool pool_;
+ ArenaAllocator allocator_;
+ CompilerOptions opts_;
+ std::unique_ptr<const InstructionSetFeatures> isa_features_;
+ HGraph* graph_;
+ std::unique_ptr<CodeGenerator> code_gen_;
+ ArenaVector<HBasicBlock*> blocks_;
+ InternalCodeAllocator code_allocator_;
};
-#define TEST_ISA(isa) \
- TEST_F(OptimizingCFITest, isa) { \
- std::vector<uint8_t> expected_asm(expected_asm_##isa, \
- expected_asm_##isa + arraysize(expected_asm_##isa)); \
- std::vector<uint8_t> expected_cfi(expected_cfi_##isa, \
- expected_cfi_##isa + arraysize(expected_cfi_##isa)); \
- TestImpl(isa, #isa, expected_asm, expected_cfi); \
+#define TEST_ISA(isa) \
+ TEST_F(OptimizingCFITest, isa) { \
+ std::vector<uint8_t> expected_asm( \
+ expected_asm_##isa, \
+ expected_asm_##isa + arraysize(expected_asm_##isa)); \
+ std::vector<uint8_t> expected_cfi( \
+ expected_cfi_##isa, \
+ expected_cfi_##isa + arraysize(expected_cfi_##isa)); \
+ TestImpl(isa, #isa, expected_asm, expected_cfi); \
}
TEST_ISA(kThumb2)
TEST_ISA(kArm64)
TEST_ISA(kX86)
TEST_ISA(kX86_64)
+TEST_ISA(kMips)
+TEST_ISA(kMips64)
+
+TEST_F(OptimizingCFITest, kThumb2Adjust) {
+ std::vector<uint8_t> expected_asm(
+ expected_asm_kThumb2_adjust,
+ expected_asm_kThumb2_adjust + arraysize(expected_asm_kThumb2_adjust));
+ std::vector<uint8_t> expected_cfi(
+ expected_cfi_kThumb2_adjust,
+ expected_cfi_kThumb2_adjust + arraysize(expected_cfi_kThumb2_adjust));
+ SetUpFrame(kThumb2);
+#define __ down_cast<arm::Thumb2Assembler*>(GetCodeGenerator()->GetAssembler())->
+ Label target;
+ __ CompareAndBranchIfZero(arm::R0, &target);
+ // Push the target out of range of CBZ.
+ for (size_t i = 0; i != 65; ++i) {
+ __ ldr(arm::R0, arm::Address(arm::R0));
+ }
+ __ Bind(&target);
+#undef __
+ Finish();
+ Check(kThumb2, "kThumb2_adjust", expected_asm, expected_cfi);
+}
+
+TEST_F(OptimizingCFITest, kMipsAdjust) {
+ // One NOP in delay slot, 1 << 15 NOPS have size 1 << 17 which exceeds 18-bit signed maximum.
+ static constexpr size_t kNumNops = 1u + (1u << 15);
+ std::vector<uint8_t> expected_asm(
+ expected_asm_kMips_adjust_head,
+ expected_asm_kMips_adjust_head + arraysize(expected_asm_kMips_adjust_head));
+ expected_asm.resize(expected_asm.size() + kNumNops * 4u, 0u);
+ expected_asm.insert(
+ expected_asm.end(),
+ expected_asm_kMips_adjust_tail,
+ expected_asm_kMips_adjust_tail + arraysize(expected_asm_kMips_adjust_tail));
+ std::vector<uint8_t> expected_cfi(
+ expected_cfi_kMips_adjust,
+ expected_cfi_kMips_adjust + arraysize(expected_cfi_kMips_adjust));
+ SetUpFrame(kMips);
+#define __ down_cast<mips::MipsAssembler*>(GetCodeGenerator()->GetAssembler())->
+ mips::MipsLabel target;
+ __ Beqz(mips::A0, &target);
+ // Push the target out of range of BEQZ.
+ for (size_t i = 0; i != kNumNops; ++i) {
+ __ Nop();
+ }
+ __ Bind(&target);
+#undef __
+ Finish();
+ Check(kMips, "kMips_adjust", expected_asm, expected_cfi);
+}
#endif // __ANDROID__
diff --git a/compiler/optimizing/optimizing_cfi_test_expected.inc b/compiler/optimizing/optimizing_cfi_test_expected.inc
index 2c2c55f295..4571ebf2d4 100644
--- a/compiler/optimizing/optimizing_cfi_test_expected.inc
+++ b/compiler/optimizing/optimizing_cfi_test_expected.inc
@@ -138,3 +138,278 @@ static constexpr uint8_t expected_cfi_kX86_64[] = {
// 0x0000002c: ret
// 0x0000002d: .cfi_restore_state
// 0x0000002d: .cfi_def_cfa_offset: 64
+
+static constexpr uint8_t expected_asm_kMips[] = {
+ 0xE4, 0xFF, 0xBD, 0x27, 0x18, 0x00, 0xBF, 0xAF, 0x14, 0x00, 0xB1, 0xAF,
+ 0x10, 0x00, 0xB0, 0xAF, 0x08, 0x00, 0xB6, 0xE7, 0x0C, 0x00, 0xB7, 0xE7,
+ 0x00, 0x00, 0xB4, 0xE7, 0x04, 0x00, 0xB5, 0xE7, 0xDC, 0xFF, 0xBD, 0x27,
+ 0x00, 0x00, 0xA4, 0xAF, 0x24, 0x00, 0xBD, 0x27, 0x00, 0x00, 0xB4, 0xC7,
+ 0x04, 0x00, 0xB5, 0xC7, 0x08, 0x00, 0xB6, 0xC7, 0x0C, 0x00, 0xB7, 0xC7,
+ 0x10, 0x00, 0xB0, 0x8F, 0x14, 0x00, 0xB1, 0x8F, 0x18, 0x00, 0xBF, 0x8F,
+ 0x1C, 0x00, 0xBD, 0x27, 0x09, 0x00, 0xE0, 0x03, 0x00, 0x00, 0x00, 0x00,
+};
+static constexpr uint8_t expected_cfi_kMips[] = {
+ 0x44, 0x0E, 0x1C, 0x44, 0x9F, 0x01, 0x44, 0x91, 0x02, 0x44, 0x90, 0x03,
+ 0x54, 0x0E, 0x40, 0x44, 0x0A, 0x44, 0x0E, 0x1C, 0x54, 0xD0, 0x44, 0xD1,
+ 0x44, 0xDF, 0x44, 0x0E, 0x00, 0x48, 0x0B, 0x0E, 0x40,
+};
+// 0x00000000: addiu r29, r29, -28
+// 0x00000004: .cfi_def_cfa_offset: 28
+// 0x00000004: sw r31, +24(r29)
+// 0x00000008: .cfi_offset: r31 at cfa-4
+// 0x00000008: sw r17, +20(r29)
+// 0x0000000c: .cfi_offset: r17 at cfa-8
+// 0x0000000c: sw r16, +16(r29)
+// 0x00000010: .cfi_offset: r16 at cfa-12
+// 0x00000010: swc1 f22, +8(r29)
+// 0x00000014: swc1 f23, +12(r29)
+// 0x00000018: swc1 f20, +0(r29)
+// 0x0000001c: swc1 f21, +4(r29)
+// 0x00000020: addiu r29, r29, -36
+// 0x00000024: .cfi_def_cfa_offset: 64
+// 0x00000024: sw r4, +0(r29)
+// 0x00000028: .cfi_remember_state
+// 0x00000028: addiu r29, r29, 36
+// 0x0000002c: .cfi_def_cfa_offset: 28
+// 0x0000002c: lwc1 f20, +0(r29)
+// 0x00000030: lwc1 f21, +4(r29)
+// 0x00000034: lwc1 f22, +8(r29)
+// 0x00000038: lwc1 f23, +12(r29)
+// 0x0000003c: lw r16, +16(r29)
+// 0x00000040: .cfi_restore: r16
+// 0x00000040: lw r17, +20(r29)
+// 0x00000044: .cfi_restore: r17
+// 0x00000044: lw r31, +24(r29)
+// 0x00000048: .cfi_restore: r31
+// 0x00000048: addiu r29, r29, 28
+// 0x0000004c: .cfi_def_cfa_offset: 0
+// 0x0000004c: jr r31
+// 0x00000050: nop
+// 0x00000054: .cfi_restore_state
+// 0x00000054: .cfi_def_cfa_offset: 64
+
+static constexpr uint8_t expected_asm_kMips64[] = {
+ 0xD8, 0xFF, 0xBD, 0x67, 0x20, 0x00, 0xBF, 0xFF, 0x18, 0x00, 0xB1, 0xFF,
+ 0x10, 0x00, 0xB0, 0xFF, 0x08, 0x00, 0xB9, 0xF7, 0x00, 0x00, 0xB8, 0xF7,
+ 0xE8, 0xFF, 0xBD, 0x67, 0x00, 0x00, 0xA4, 0xFF, 0x18, 0x00, 0xBD, 0x67,
+ 0x00, 0x00, 0xB8, 0xD7, 0x08, 0x00, 0xB9, 0xD7, 0x10, 0x00, 0xB0, 0xDF,
+ 0x18, 0x00, 0xB1, 0xDF, 0x20, 0x00, 0xBF, 0xDF, 0x28, 0x00, 0xBD, 0x67,
+ 0x09, 0x00, 0xE0, 0x03, 0x00, 0x00, 0x00, 0x00,
+};
+static constexpr uint8_t expected_cfi_kMips64[] = {
+ 0x44, 0x0E, 0x28, 0x44, 0x9F, 0x02, 0x44, 0x91, 0x04, 0x44, 0x90, 0x06,
+ 0x4C, 0x0E, 0x40, 0x44, 0x0A, 0x44, 0x0E, 0x28, 0x4C, 0xD0, 0x44, 0xD1,
+ 0x44, 0xDF, 0x44, 0x0E, 0x00, 0x48, 0x0B, 0x0E, 0x40,
+};
+// 0x00000000: daddiu r29, r29, -40
+// 0x00000004: .cfi_def_cfa_offset: 40
+// 0x00000004: sd r31, +32(r29)
+// 0x00000008: .cfi_offset: r31 at cfa-8
+// 0x00000008: sd r17, +24(r29)
+// 0x0000000c: .cfi_offset: r17 at cfa-16
+// 0x0000000c: sd r16, +16(r29)
+// 0x00000010: .cfi_offset: r16 at cfa-24
+// 0x00000010: sdc1 f25, +8(r29)
+// 0x00000014: sdc1 f24, +0(r29)
+// 0x00000018: daddiu r29, r29, -24
+// 0x0000001c: .cfi_def_cfa_offset: 64
+// 0x0000001c: sd r4, +0(r29)
+// 0x00000020: .cfi_remember_state
+// 0x00000020: daddiu r29, r29, 24
+// 0x00000024: .cfi_def_cfa_offset: 40
+// 0x00000024: ldc1 f24, +0(r29)
+// 0x00000028: ldc1 f25, +8(r29)
+// 0x0000002c: ld r16, +16(r29)
+// 0x00000030: .cfi_restore: r16
+// 0x00000030: ld r17, +24(r29)
+// 0x00000034: .cfi_restore: r17
+// 0x00000034: ld r31, +32(r29)
+// 0x00000038: .cfi_restore: r31
+// 0x00000038: daddiu r29, r29, 40
+// 0x0000003c: .cfi_def_cfa_offset: 0
+// 0x0000003c: jr r31
+// 0x00000040: nop
+// 0x00000044: .cfi_restore_state
+// 0x00000044: .cfi_def_cfa_offset: 64
+
+static constexpr uint8_t expected_asm_kThumb2_adjust[] = {
+ 0x60, 0xB5, 0x2D, 0xED, 0x02, 0x8A, 0x8B, 0xB0, 0x00, 0x90, 0x00, 0x28,
+ 0x40, 0xD0, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68,
+ 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68,
+ 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68,
+ 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68,
+ 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68,
+ 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68,
+ 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68,
+ 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68,
+ 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68,
+ 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68,
+ 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68, 0x00, 0x68,
+ 0x0B, 0xB0, 0xBD, 0xEC, 0x02, 0x8A, 0x60, 0xBD,
+};
+static constexpr uint8_t expected_cfi_kThumb2_adjust[] = {
+ 0x42, 0x0E, 0x0C, 0x85, 0x03, 0x86, 0x02, 0x8E, 0x01, 0x44, 0x0E, 0x14,
+ 0x05, 0x50, 0x05, 0x05, 0x51, 0x04, 0x42, 0x0E, 0x40, 0x02, 0x88, 0x0A,
+ 0x42, 0x0E, 0x14, 0x44, 0x0E, 0x0C, 0x06, 0x50, 0x06, 0x51, 0x42, 0x0B,
+ 0x0E, 0x40,
+};
+// 0x00000000: push {r5, r6, lr}
+// 0x00000002: .cfi_def_cfa_offset: 12
+// 0x00000002: .cfi_offset: r5 at cfa-12
+// 0x00000002: .cfi_offset: r6 at cfa-8
+// 0x00000002: .cfi_offset: r14 at cfa-4
+// 0x00000002: vpush.f32 {s16-s17}
+// 0x00000006: .cfi_def_cfa_offset: 20
+// 0x00000006: .cfi_offset_extended: r80 at cfa-20
+// 0x00000006: .cfi_offset_extended: r81 at cfa-16
+// 0x00000006: sub sp, sp, #44
+// 0x00000008: .cfi_def_cfa_offset: 64
+// 0x00000008: str r0, [sp, #0]
+// 0x0000000a: cmp r0, #0
+// 0x0000000c: beq +128 (0x00000090)
+// 0x0000000e: ldr r0, [r0, #0]
+// 0x00000010: ldr r0, [r0, #0]
+// 0x00000012: ldr r0, [r0, #0]
+// 0x00000014: ldr r0, [r0, #0]
+// 0x00000016: ldr r0, [r0, #0]
+// 0x00000018: ldr r0, [r0, #0]
+// 0x0000001a: ldr r0, [r0, #0]
+// 0x0000001c: ldr r0, [r0, #0]
+// 0x0000001e: ldr r0, [r0, #0]
+// 0x00000020: ldr r0, [r0, #0]
+// 0x00000022: ldr r0, [r0, #0]
+// 0x00000024: ldr r0, [r0, #0]
+// 0x00000026: ldr r0, [r0, #0]
+// 0x00000028: ldr r0, [r0, #0]
+// 0x0000002a: ldr r0, [r0, #0]
+// 0x0000002c: ldr r0, [r0, #0]
+// 0x0000002e: ldr r0, [r0, #0]
+// 0x00000030: ldr r0, [r0, #0]
+// 0x00000032: ldr r0, [r0, #0]
+// 0x00000034: ldr r0, [r0, #0]
+// 0x00000036: ldr r0, [r0, #0]
+// 0x00000038: ldr r0, [r0, #0]
+// 0x0000003a: ldr r0, [r0, #0]
+// 0x0000003c: ldr r0, [r0, #0]
+// 0x0000003e: ldr r0, [r0, #0]
+// 0x00000040: ldr r0, [r0, #0]
+// 0x00000042: ldr r0, [r0, #0]
+// 0x00000044: ldr r0, [r0, #0]
+// 0x00000046: ldr r0, [r0, #0]
+// 0x00000048: ldr r0, [r0, #0]
+// 0x0000004a: ldr r0, [r0, #0]
+// 0x0000004c: ldr r0, [r0, #0]
+// 0x0000004e: ldr r0, [r0, #0]
+// 0x00000050: ldr r0, [r0, #0]
+// 0x00000052: ldr r0, [r0, #0]
+// 0x00000054: ldr r0, [r0, #0]
+// 0x00000056: ldr r0, [r0, #0]
+// 0x00000058: ldr r0, [r0, #0]
+// 0x0000005a: ldr r0, [r0, #0]
+// 0x0000005c: ldr r0, [r0, #0]
+// 0x0000005e: ldr r0, [r0, #0]
+// 0x00000060: ldr r0, [r0, #0]
+// 0x00000062: ldr r0, [r0, #0]
+// 0x00000064: ldr r0, [r0, #0]
+// 0x00000066: ldr r0, [r0, #0]
+// 0x00000068: ldr r0, [r0, #0]
+// 0x0000006a: ldr r0, [r0, #0]
+// 0x0000006c: ldr r0, [r0, #0]
+// 0x0000006e: ldr r0, [r0, #0]
+// 0x00000070: ldr r0, [r0, #0]
+// 0x00000072: ldr r0, [r0, #0]
+// 0x00000074: ldr r0, [r0, #0]
+// 0x00000076: ldr r0, [r0, #0]
+// 0x00000078: ldr r0, [r0, #0]
+// 0x0000007a: ldr r0, [r0, #0]
+// 0x0000007c: ldr r0, [r0, #0]
+// 0x0000007e: ldr r0, [r0, #0]
+// 0x00000080: ldr r0, [r0, #0]
+// 0x00000082: ldr r0, [r0, #0]
+// 0x00000084: ldr r0, [r0, #0]
+// 0x00000086: ldr r0, [r0, #0]
+// 0x00000088: ldr r0, [r0, #0]
+// 0x0000008a: ldr r0, [r0, #0]
+// 0x0000008c: ldr r0, [r0, #0]
+// 0x0000008e: ldr r0, [r0, #0]
+// 0x00000090: .cfi_remember_state
+// 0x00000090: add sp, sp, #44
+// 0x00000092: .cfi_def_cfa_offset: 20
+// 0x00000092: vpop.f32 {s16-s17}
+// 0x00000096: .cfi_def_cfa_offset: 12
+// 0x00000096: .cfi_restore_extended: r80
+// 0x00000096: .cfi_restore_extended: r81
+// 0x00000096: pop {r5, r6, pc}
+// 0x00000098: .cfi_restore_state
+// 0x00000098: .cfi_def_cfa_offset: 64
+
+static constexpr uint8_t expected_asm_kMips_adjust_head[] = {
+ 0xE4, 0xFF, 0xBD, 0x27, 0x18, 0x00, 0xBF, 0xAF, 0x14, 0x00, 0xB1, 0xAF,
+ 0x10, 0x00, 0xB0, 0xAF, 0x08, 0x00, 0xB6, 0xE7, 0x0C, 0x00, 0xB7, 0xE7,
+ 0x00, 0x00, 0xB4, 0xE7, 0x04, 0x00, 0xB5, 0xE7, 0xDC, 0xFF, 0xBD, 0x27,
+ 0x00, 0x00, 0xA4, 0xAF, 0x08, 0x00, 0x04, 0x14, 0xFC, 0xFF, 0xBD, 0x27,
+ 0x00, 0x00, 0xBF, 0xAF, 0x00, 0x00, 0x10, 0x04, 0x02, 0x00, 0x01, 0x3C,
+ 0x18, 0x00, 0x21, 0x34, 0x21, 0x08, 0x3F, 0x00, 0x00, 0x00, 0xBF, 0x8F,
+ 0x09, 0x00, 0x20, 0x00, 0x04, 0x00, 0xBD, 0x27,
+};
+static constexpr uint8_t expected_asm_kMips_adjust_tail[] = {
+ 0x24, 0x00, 0xBD, 0x27, 0x00, 0x00, 0xB4, 0xC7, 0x04, 0x00, 0xB5, 0xC7,
+ 0x08, 0x00, 0xB6, 0xC7, 0x0C, 0x00, 0xB7, 0xC7, 0x10, 0x00, 0xB0, 0x8F,
+ 0x14, 0x00, 0xB1, 0x8F, 0x18, 0x00, 0xBF, 0x8F, 0x1C, 0x00, 0xBD, 0x27,
+ 0x09, 0x00, 0xE0, 0x03, 0x00, 0x00, 0x00, 0x00,
+};
+static constexpr uint8_t expected_cfi_kMips_adjust[] = {
+ 0x44, 0x0E, 0x1C, 0x44, 0x9F, 0x01, 0x44, 0x91, 0x02, 0x44, 0x90, 0x03,
+ 0x54, 0x0E, 0x40, 0x4C, 0x0E, 0x44, 0x60, 0x0E, 0x40, 0x04, 0x04, 0x00,
+ 0x02, 0x00, 0x0A, 0x44, 0x0E, 0x1C, 0x54, 0xD0, 0x44, 0xD1, 0x44, 0xDF,
+ 0x44, 0x0E, 0x00, 0x48, 0x0B, 0x0E, 0x40,
+};
+// 0x00000000: addiu r29, r29, -28
+// 0x00000004: .cfi_def_cfa_offset: 28
+// 0x00000004: sw r31, +24(r29)
+// 0x00000008: .cfi_offset: r31 at cfa-4
+// 0x00000008: sw r17, +20(r29)
+// 0x0000000c: .cfi_offset: r17 at cfa-8
+// 0x0000000c: sw r16, +16(r29)
+// 0x00000010: .cfi_offset: r16 at cfa-12
+// 0x00000010: swc1 f22, +8(r29)
+// 0x00000014: swc1 f23, +12(r29)
+// 0x00000018: swc1 f20, +0(r29)
+// 0x0000001c: swc1 f21, +4(r29)
+// 0x00000020: addiu r29, r29, -36
+// 0x00000024: .cfi_def_cfa_offset: 64
+// 0x00000024: sw r4, +0(r29)
+// 0x00000028: bne r0, r4, 0x0000004c ; +36
+// 0x0000002c: addiu r29, r29, -4
+// 0x00000030: .cfi_def_cfa_offset: 68
+// 0x00000030: sw r31, +0(r29)
+// 0x00000034: bltzal r0, 0x00000038 ; +4
+// 0x00000038: lui r1, 0x20000
+// 0x0000003c: ori r1, r1, 24
+// 0x00000040: addu r1, r1, r31
+// 0x00000044: lw r31, +0(r29)
+// 0x00000048: jr r1
+// 0x0000004c: addiu r29, r29, 4
+// 0x00000050: .cfi_def_cfa_offset: 64
+// 0x00000050: nop
+// ...
+// 0x00020050: nop
+// 0x00020054: .cfi_remember_state
+// 0x00020054: addiu r29, r29, 36
+// 0x00020058: .cfi_def_cfa_offset: 28
+// 0x00020058: lwc1 f20, +0(r29)
+// 0x0002005c: lwc1 f21, +4(r29)
+// 0x00020060: lwc1 f22, +8(r29)
+// 0x00020064: lwc1 f23, +12(r29)
+// 0x00020068: lw r16, +16(r29)
+// 0x0002006c: .cfi_restore: r16
+// 0x0002006c: lw r17, +20(r29)
+// 0x00020070: .cfi_restore: r17
+// 0x00020070: lw r31, +24(r29)
+// 0x00020074: .cfi_restore: r31
+// 0x00020074: addiu r29, r29, 28
+// 0x00020078: .cfi_def_cfa_offset: 0
+// 0x00020078: jr r31
+// 0x0002007c: nop
+// 0x00020080: .cfi_restore_state
+// 0x00020080: .cfi_def_cfa_offset: 64
diff --git a/compiler/optimizing/optimizing_compiler.cc b/compiler/optimizing/optimizing_compiler.cc
index 8cb2cfc816..7e3c5e602e 100644
--- a/compiler/optimizing/optimizing_compiler.cc
+++ b/compiler/optimizing/optimizing_compiler.cc
@@ -56,6 +56,7 @@
#include "inliner.h"
#include "instruction_simplifier.h"
#include "intrinsics.h"
+#include "jit/jit_code_cache.h"
#include "licm.h"
#include "jni/quick/jni_compiler.h"
#include "load_store_elimination.h"
@@ -258,15 +259,6 @@ class OptimizingCompiler FINAL : public Compiler {
const DexFile& dex_file,
Handle<mirror::DexCache> dex_cache) const OVERRIDE;
- CompiledMethod* TryCompile(const DexFile::CodeItem* code_item,
- uint32_t access_flags,
- InvokeType invoke_type,
- uint16_t class_def_idx,
- uint32_t method_idx,
- jobject class_loader,
- const DexFile& dex_file,
- Handle<mirror::DexCache> dex_cache) const;
-
CompiledMethod* JniCompile(uint32_t access_flags,
uint32_t method_idx,
const DexFile& dex_file) const OVERRIDE {
@@ -291,23 +283,45 @@ class OptimizingCompiler FINAL : public Compiler {
}
}
+ bool JitCompile(Thread* self, jit::JitCodeCache* code_cache, ArtMethod* method)
+ OVERRIDE
+ SHARED_REQUIRES(Locks::mutator_lock_);
+
private:
// Whether we should run any optimization or register allocation. If false, will
// just run the code generation after the graph was built.
const bool run_optimizations_;
- // Optimize and compile `graph`.
- CompiledMethod* CompileOptimized(HGraph* graph,
- CodeGenerator* codegen,
- CompilerDriver* driver,
- const DexCompilationUnit& dex_compilation_unit,
- PassObserver* pass_observer) const;
-
- // Just compile without doing optimizations.
- CompiledMethod* CompileBaseline(CodeGenerator* codegen,
- CompilerDriver* driver,
- const DexCompilationUnit& dex_compilation_unit,
- PassObserver* pass_observer) const;
+ // Create a 'CompiledMethod' for an optimized graph.
+ CompiledMethod* EmitOptimized(ArenaAllocator* arena,
+ CodeVectorAllocator* code_allocator,
+ CodeGenerator* codegen,
+ CompilerDriver* driver) const;
+
+ // Create a 'CompiledMethod' for a non-optimized graph.
+ CompiledMethod* EmitBaseline(ArenaAllocator* arena,
+ CodeVectorAllocator* code_allocator,
+ CodeGenerator* codegen,
+ CompilerDriver* driver) const;
+
+ // Try compiling a method and return the code generator used for
+ // compiling it.
+ // This method:
+ // 1) Builds the graph. Returns null if it failed to build it.
+ // 2) If `run_optimizations_` is set:
+ // 2.1) Transform the graph to SSA. Returns null if it failed.
+ // 2.2) Run optimizations on the graph, including register allocator.
+ // 3) Generate code with the `code_allocator` provided.
+ CodeGenerator* TryCompile(ArenaAllocator* arena,
+ CodeVectorAllocator* code_allocator,
+ const DexFile::CodeItem* code_item,
+ uint32_t access_flags,
+ InvokeType invoke_type,
+ uint16_t class_def_idx,
+ uint32_t method_idx,
+ jobject class_loader,
+ const DexFile& dex_file,
+ Handle<mirror::DexCache> dex_cache) const;
std::unique_ptr<OptimizingCompilerStats> compilation_stats_;
@@ -446,13 +460,32 @@ static void RunArchOptimizations(InstructionSet instruction_set,
}
}
+NO_INLINE // Avoid increasing caller's frame size by large stack-allocated objects.
+static void AllocateRegisters(HGraph* graph,
+ CodeGenerator* codegen,
+ PassObserver* pass_observer) {
+ PrepareForRegisterAllocation(graph).Run();
+ SsaLivenessAnalysis liveness(graph, codegen);
+ {
+ PassScope scope(SsaLivenessAnalysis::kLivenessPassName, pass_observer);
+ liveness.Analyze();
+ }
+ {
+ PassScope scope(RegisterAllocator::kRegisterAllocatorPassName, pass_observer);
+ RegisterAllocator(graph->GetArena(), codegen, liveness).AllocateRegisters();
+ }
+}
+
static void RunOptimizations(HGraph* graph,
CodeGenerator* codegen,
CompilerDriver* driver,
OptimizingCompilerStats* stats,
const DexCompilationUnit& dex_compilation_unit,
- PassObserver* pass_observer,
- StackHandleScopeCollection* handles) {
+ PassObserver* pass_observer) {
+ ScopedObjectAccess soa(Thread::Current());
+ StackHandleScopeCollection handles(soa.Self());
+ ScopedThreadSuspension sts(soa.Self(), kNative);
+
ArenaAllocator* arena = graph->GetArena();
HDeadCodeElimination* dce1 = new (arena) HDeadCodeElimination(
graph, stats, HDeadCodeElimination::kInitialDeadCodeEliminationPassName);
@@ -469,7 +502,7 @@ static void RunOptimizations(HGraph* graph,
HInductionVarAnalysis* induction = new (arena) HInductionVarAnalysis(graph);
BoundsCheckElimination* bce = new (arena) BoundsCheckElimination(graph, induction);
ReferenceTypePropagation* type_propagation =
- new (arena) ReferenceTypePropagation(graph, handles);
+ new (arena) ReferenceTypePropagation(graph, &handles);
HSharpening* sharpening = new (arena) HSharpening(graph, codegen, dex_compilation_unit, driver);
InstructionSimplifier* simplify2 = new (arena) InstructionSimplifier(
graph, stats, "instruction_simplifier_after_types");
@@ -492,7 +525,7 @@ static void RunOptimizations(HGraph* graph,
RunOptimizations(optimizations1, arraysize(optimizations1), pass_observer);
- MaybeRunInliner(graph, codegen, driver, stats, dex_compilation_unit, pass_observer, handles);
+ MaybeRunInliner(graph, codegen, driver, stats, dex_compilation_unit, pass_observer, &handles);
// TODO: Update passes incompatible with try/catch so we have the same
// pipeline for all methods.
@@ -532,6 +565,7 @@ static void RunOptimizations(HGraph* graph,
}
RunArchOptimizations(driver->GetInstructionSet(), graph, stats, pass_observer);
+ AllocateRegisters(graph, codegen, pass_observer);
}
// The stack map we generate must be 4-byte aligned on ARM. Since existing
@@ -545,22 +579,6 @@ static ArrayRef<const uint8_t> AlignVectorSize(ArenaVector<uint8_t>& vector) {
return ArrayRef<const uint8_t>(vector);
}
-NO_INLINE // Avoid increasing caller's frame size by large stack-allocated objects.
-static void AllocateRegisters(HGraph* graph,
- CodeGenerator* codegen,
- PassObserver* pass_observer) {
- PrepareForRegisterAllocation(graph).Run();
- SsaLivenessAnalysis liveness(graph, codegen);
- {
- PassScope scope(SsaLivenessAnalysis::kLivenessPassName, pass_observer);
- liveness.Analyze();
- }
- {
- PassScope scope(RegisterAllocator::kRegisterAllocatorPassName, pass_observer);
- RegisterAllocator(graph->GetArena(), codegen, liveness).AllocateRegisters();
- }
-}
-
static ArenaVector<LinkerPatch> EmitAndSortLinkerPatches(CodeGenerator* codegen) {
ArenaVector<LinkerPatch> linker_patches(codegen->GetGraph()->GetArena()->Adapter());
codegen->EmitLinkerPatches(&linker_patches);
@@ -574,74 +592,42 @@ static ArenaVector<LinkerPatch> EmitAndSortLinkerPatches(CodeGenerator* codegen)
return linker_patches;
}
-CompiledMethod* OptimizingCompiler::CompileOptimized(HGraph* graph,
- CodeGenerator* codegen,
- CompilerDriver* compiler_driver,
- const DexCompilationUnit& dex_compilation_unit,
- PassObserver* pass_observer) const {
- ScopedObjectAccess soa(Thread::Current());
- StackHandleScopeCollection handles(soa.Self());
- soa.Self()->TransitionFromRunnableToSuspended(kNative);
- RunOptimizations(graph,
- codegen,
- compiler_driver,
- compilation_stats_.get(),
- dex_compilation_unit,
- pass_observer,
- &handles);
-
- AllocateRegisters(graph, codegen, pass_observer);
-
- ArenaAllocator* arena = graph->GetArena();
- CodeVectorAllocator allocator(arena);
- DefaultSrcMap src_mapping_table;
- codegen->SetSrcMap(compiler_driver->GetCompilerOptions().GetGenerateDebugInfo()
- ? &src_mapping_table
- : nullptr);
- codegen->CompileOptimized(&allocator);
-
+CompiledMethod* OptimizingCompiler::EmitOptimized(ArenaAllocator* arena,
+ CodeVectorAllocator* code_allocator,
+ CodeGenerator* codegen,
+ CompilerDriver* compiler_driver) const {
ArenaVector<LinkerPatch> linker_patches = EmitAndSortLinkerPatches(codegen);
-
ArenaVector<uint8_t> stack_map(arena->Adapter(kArenaAllocStackMaps));
- codegen->BuildStackMaps(&stack_map);
+ stack_map.resize(codegen->ComputeStackMapsSize());
+ codegen->BuildStackMaps(MemoryRegion(stack_map.data(), stack_map.size()));
MaybeRecordStat(MethodCompilationStat::kCompiledOptimized);
CompiledMethod* compiled_method = CompiledMethod::SwapAllocCompiledMethod(
compiler_driver,
codegen->GetInstructionSet(),
- ArrayRef<const uint8_t>(allocator.GetMemory()),
+ ArrayRef<const uint8_t>(code_allocator->GetMemory()),
// Follow Quick's behavior and set the frame size to zero if it is
// considered "empty" (see the definition of
// art::CodeGenerator::HasEmptyFrame).
codegen->HasEmptyFrame() ? 0 : codegen->GetFrameSize(),
codegen->GetCoreSpillMask(),
codegen->GetFpuSpillMask(),
- ArrayRef<const SrcMapElem>(src_mapping_table),
+ ArrayRef<const SrcMapElem>(codegen->GetSrcMappingTable()),
ArrayRef<const uint8_t>(), // mapping_table.
ArrayRef<const uint8_t>(stack_map),
ArrayRef<const uint8_t>(), // native_gc_map.
ArrayRef<const uint8_t>(*codegen->GetAssembler()->cfi().data()),
ArrayRef<const LinkerPatch>(linker_patches));
- pass_observer->DumpDisassembly();
- soa.Self()->TransitionFromSuspendedToRunnable();
return compiled_method;
}
-CompiledMethod* OptimizingCompiler::CompileBaseline(
+CompiledMethod* OptimizingCompiler::EmitBaseline(
+ ArenaAllocator* arena,
+ CodeVectorAllocator* code_allocator,
CodeGenerator* codegen,
- CompilerDriver* compiler_driver,
- const DexCompilationUnit& dex_compilation_unit,
- PassObserver* pass_observer) const {
- ArenaAllocator* arena = codegen->GetGraph()->GetArena();
- CodeVectorAllocator allocator(arena);
- DefaultSrcMap src_mapping_table;
- codegen->SetSrcMap(compiler_driver->GetCompilerOptions().GetGenerateDebugInfo()
- ? &src_mapping_table
- : nullptr);
- codegen->CompileBaseline(&allocator);
-
+ CompilerDriver* compiler_driver) const {
ArenaVector<LinkerPatch> linker_patches = EmitAndSortLinkerPatches(codegen);
ArenaVector<uint8_t> mapping_table(arena->Adapter(kArenaAllocBaselineMaps));
@@ -649,37 +635,38 @@ CompiledMethod* OptimizingCompiler::CompileBaseline(
ArenaVector<uint8_t> vmap_table(arena->Adapter(kArenaAllocBaselineMaps));
codegen->BuildVMapTable(&vmap_table);
ArenaVector<uint8_t> gc_map(arena->Adapter(kArenaAllocBaselineMaps));
- codegen->BuildNativeGCMap(&gc_map, dex_compilation_unit);
+ codegen->BuildNativeGCMap(&gc_map, *compiler_driver);
MaybeRecordStat(MethodCompilationStat::kCompiledBaseline);
CompiledMethod* compiled_method = CompiledMethod::SwapAllocCompiledMethod(
compiler_driver,
codegen->GetInstructionSet(),
- ArrayRef<const uint8_t>(allocator.GetMemory()),
+ ArrayRef<const uint8_t>(code_allocator->GetMemory()),
// Follow Quick's behavior and set the frame size to zero if it is
// considered "empty" (see the definition of
// art::CodeGenerator::HasEmptyFrame).
codegen->HasEmptyFrame() ? 0 : codegen->GetFrameSize(),
codegen->GetCoreSpillMask(),
codegen->GetFpuSpillMask(),
- ArrayRef<const SrcMapElem>(src_mapping_table),
+ ArrayRef<const SrcMapElem>(codegen->GetSrcMappingTable()),
AlignVectorSize(mapping_table),
AlignVectorSize(vmap_table),
AlignVectorSize(gc_map),
ArrayRef<const uint8_t>(*codegen->GetAssembler()->cfi().data()),
ArrayRef<const LinkerPatch>(linker_patches));
- pass_observer->DumpDisassembly();
return compiled_method;
}
-CompiledMethod* OptimizingCompiler::TryCompile(const DexFile::CodeItem* code_item,
- uint32_t access_flags,
- InvokeType invoke_type,
- uint16_t class_def_idx,
- uint32_t method_idx,
- jobject class_loader,
- const DexFile& dex_file,
- Handle<mirror::DexCache> dex_cache) const {
+CodeGenerator* OptimizingCompiler::TryCompile(ArenaAllocator* arena,
+ CodeVectorAllocator* code_allocator,
+ const DexFile::CodeItem* code_item,
+ uint32_t access_flags,
+ InvokeType invoke_type,
+ uint16_t class_def_idx,
+ uint32_t method_idx,
+ jobject class_loader,
+ const DexFile& dex_file,
+ Handle<mirror::DexCache> dex_cache) const {
std::string method_name = PrettyMethod(method_idx, dex_file);
MaybeRecordStat(MethodCompilationStat::kAttemptCompilation);
CompilerDriver* compiler_driver = GetCompilerDriver();
@@ -721,13 +708,10 @@ CompiledMethod* OptimizingCompiler::TryCompile(const DexFile::CodeItem* code_ite
&& compiler_driver->RequiresConstructorBarrier(Thread::Current(),
dex_compilation_unit.GetDexFile(),
dex_compilation_unit.GetClassDefIndex());
- ArenaAllocator arena(Runtime::Current()->GetArenaPool());
- HGraph* graph = new (&arena) HGraph(
- &arena, dex_file, method_idx, requires_barrier, compiler_driver->GetInstructionSet(),
+ HGraph* graph = new (arena) HGraph(
+ arena, dex_file, method_idx, requires_barrier, compiler_driver->GetInstructionSet(),
kInvalidInvokeType, compiler_driver->GetCompilerOptions().GetDebuggable());
- bool shouldOptimize = method_name.find("$opt$reg$") != std::string::npos && run_optimizations_;
-
std::unique_ptr<CodeGenerator> codegen(
CodeGenerator::Create(graph,
instruction_set,
@@ -779,16 +763,8 @@ CompiledMethod* OptimizingCompiler::TryCompile(const DexFile::CodeItem* code_ite
}
}
- bool can_allocate_registers = RegisterAllocator::CanAllocateRegistersFor(*graph, instruction_set);
-
- // `run_optimizations_` is set explicitly (either through a compiler filter
- // or the debuggable flag). If it is set, we can run baseline. Otherwise, we fall back
- // to Quick.
- bool can_use_baseline = !run_optimizations_ && builder.CanUseBaselineForStringInit();
- CompiledMethod* compiled_method = nullptr;
- if (run_optimizations_ && can_allocate_registers) {
- VLOG(compiler) << "Optimizing " << method_name;
-
+ VLOG(compiler) << "Optimizing " << method_name;
+ if (run_optimizations_) {
{
PassScope scope(SsaBuilder::kSsaBuilderPassName, &pass_observer);
if (!graph->TryBuildingSsa()) {
@@ -800,37 +776,26 @@ CompiledMethod* OptimizingCompiler::TryCompile(const DexFile::CodeItem* code_ite
}
}
- compiled_method = CompileOptimized(graph,
- codegen.get(),
- compiler_driver,
- dex_compilation_unit,
- &pass_observer);
- } else if (shouldOptimize && can_allocate_registers) {
- LOG(FATAL) << "Could not allocate registers in optimizing compiler";
- UNREACHABLE();
- } else if (can_use_baseline) {
- VLOG(compiler) << "Compile baseline " << method_name;
-
- if (!run_optimizations_) {
- MaybeRecordStat(MethodCompilationStat::kNotOptimizedDisabled);
- } else if (!can_allocate_registers) {
- MaybeRecordStat(MethodCompilationStat::kNotOptimizedRegisterAllocator);
- }
-
- compiled_method = CompileBaseline(codegen.get(),
- compiler_driver,
- dex_compilation_unit,
- &pass_observer);
+ RunOptimizations(graph,
+ codegen.get(),
+ compiler_driver,
+ compilation_stats_.get(),
+ dex_compilation_unit,
+ &pass_observer);
+ codegen->CompileOptimized(code_allocator);
+ } else {
+ codegen->CompileBaseline(code_allocator);
}
+ pass_observer.DumpDisassembly();
if (kArenaAllocatorCountAllocations) {
- if (arena.BytesAllocated() > 4 * MB) {
- MemStats mem_stats(arena.GetMemStats());
+ if (arena->BytesAllocated() > 4 * MB) {
+ MemStats mem_stats(arena->GetMemStats());
LOG(INFO) << PrettyMethod(method_idx, dex_file) << " " << Dumpable<MemStats>(mem_stats);
}
}
- return compiled_method;
+ return codegen.release();
}
static bool CanHandleVerificationFailure(const VerifiedMethod* verified_method) {
@@ -852,26 +817,37 @@ CompiledMethod* OptimizingCompiler::Compile(const DexFile::CodeItem* code_item,
Handle<mirror::DexCache> dex_cache) const {
CompilerDriver* compiler_driver = GetCompilerDriver();
CompiledMethod* method = nullptr;
- if (Runtime::Current()->IsAotCompiler()) {
- const VerifiedMethod* verified_method = compiler_driver->GetVerifiedMethod(&dex_file, method_idx);
- DCHECK(!verified_method->HasRuntimeThrow());
- if (compiler_driver->IsMethodVerifiedWithoutFailures(method_idx, class_def_idx, dex_file)
- || CanHandleVerificationFailure(verified_method)) {
- method = TryCompile(code_item, access_flags, invoke_type, class_def_idx,
- method_idx, jclass_loader, dex_file, dex_cache);
- } else {
- if (compiler_driver->GetCompilerOptions().VerifyAtRuntime()) {
- MaybeRecordStat(MethodCompilationStat::kNotCompiledVerifyAtRuntime);
+ DCHECK(Runtime::Current()->IsAotCompiler());
+ const VerifiedMethod* verified_method = compiler_driver->GetVerifiedMethod(&dex_file, method_idx);
+ DCHECK(!verified_method->HasRuntimeThrow());
+ if (compiler_driver->IsMethodVerifiedWithoutFailures(method_idx, class_def_idx, dex_file)
+ || CanHandleVerificationFailure(verified_method)) {
+ ArenaAllocator arena(Runtime::Current()->GetArenaPool());
+ CodeVectorAllocator code_allocator(&arena);
+ std::unique_ptr<CodeGenerator> codegen(
+ TryCompile(&arena,
+ &code_allocator,
+ code_item,
+ access_flags,
+ invoke_type,
+ class_def_idx,
+ method_idx,
+ jclass_loader,
+ dex_file,
+ dex_cache));
+ if (codegen.get() != nullptr) {
+ if (run_optimizations_) {
+ method = EmitOptimized(&arena, &code_allocator, codegen.get(), compiler_driver);
} else {
- MaybeRecordStat(MethodCompilationStat::kNotCompiledClassNotVerified);
+ method = EmitBaseline(&arena, &code_allocator, codegen.get(), compiler_driver);
}
}
} else {
- // This is for the JIT compiler, which has already ensured the class is verified.
- // We can go straight to compiling.
- DCHECK(Runtime::Current()->UseJit());
- method = TryCompile(code_item, access_flags, invoke_type, class_def_idx,
- method_idx, jclass_loader, dex_file, dex_cache);
+ if (compiler_driver->GetCompilerOptions().VerifyAtRuntime()) {
+ MaybeRecordStat(MethodCompilationStat::kNotCompiledVerifyAtRuntime);
+ } else {
+ MaybeRecordStat(MethodCompilationStat::kNotCompiledClassNotVerified);
+ }
}
if (kIsDebugBuild &&
@@ -896,4 +872,70 @@ bool IsCompilingWithCoreImage() {
return EndsWith(image, "core.art") || EndsWith(image, "core-optimizing.art");
}
+bool OptimizingCompiler::JitCompile(Thread* self,
+ jit::JitCodeCache* code_cache,
+ ArtMethod* method) {
+ StackHandleScope<2> hs(self);
+ Handle<mirror::ClassLoader> class_loader(hs.NewHandle(
+ method->GetDeclaringClass()->GetClassLoader()));
+ Handle<mirror::DexCache> dex_cache(hs.NewHandle(method->GetDexCache()));
+
+ jobject jclass_loader = class_loader.ToJObject();
+ const DexFile* dex_file = method->GetDexFile();
+ const uint16_t class_def_idx = method->GetClassDefIndex();
+ const DexFile::CodeItem* code_item = dex_file->GetCodeItem(method->GetCodeItemOffset());
+ const uint32_t method_idx = method->GetDexMethodIndex();
+ const uint32_t access_flags = method->GetAccessFlags();
+ const InvokeType invoke_type = method->GetInvokeType();
+
+ ArenaAllocator arena(Runtime::Current()->GetArenaPool());
+ CodeVectorAllocator code_allocator(&arena);
+ std::unique_ptr<CodeGenerator> codegen;
+ {
+ // Go to native so that we don't block GC during compilation.
+ ScopedThreadSuspension sts(self, kNative);
+
+ DCHECK(run_optimizations_);
+ codegen.reset(
+ TryCompile(&arena,
+ &code_allocator,
+ code_item,
+ access_flags,
+ invoke_type,
+ class_def_idx,
+ method_idx,
+ jclass_loader,
+ *dex_file,
+ dex_cache));
+ if (codegen.get() == nullptr) {
+ return false;
+ }
+ }
+
+ size_t stack_map_size = codegen->ComputeStackMapsSize();
+ uint8_t* stack_map_data = code_cache->ReserveData(self, stack_map_size);
+ if (stack_map_data == nullptr) {
+ return false;
+ }
+ codegen->BuildStackMaps(MemoryRegion(stack_map_data, stack_map_size));
+ const void* code = code_cache->CommitCode(
+ self,
+ method,
+ nullptr,
+ stack_map_data,
+ nullptr,
+ codegen->HasEmptyFrame() ? 0 : codegen->GetFrameSize(),
+ codegen->GetCoreSpillMask(),
+ codegen->GetFpuSpillMask(),
+ code_allocator.GetMemory().data(),
+ code_allocator.GetSize());
+
+ if (code == nullptr) {
+ code_cache->ClearData(self, stack_map_data);
+ return false;
+ }
+
+ return true;
+}
+
} // namespace art
generated by cgit v1.2.3-59-g8ed1b (git 2.39.0) at 2025-11-05 09:28:56 +0000